/* i386.c -- Assemble code for the Intel 80386
Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
- 2000, 2001, 2002, 2003
+ 2000, 2001, 2002, 2003, 2004, 2005
Free Software Foundation, Inc.
This file is part of GAS, the GNU Assembler.
/* Intel 80386 machine specific gas.
Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
x86_64 support by Jan Hubicka (jh@suse.cz)
+ VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
Bugs & suggestions are completely welcome. This is free software.
Please help us make it better. */
#include "dwarf2dbg.h"
#include "dw2gencfi.h"
#include "opcode/i386.h"
+#include "elf/x86-64.h"
#ifndef REGISTER_WARNINGS
#define REGISTER_WARNINGS 1
#define SCALE1_WHEN_NO_INDEX 1
#endif
-#ifdef BFD_ASSEMBLER
-#define RELOC_ENUM enum bfd_reloc_code_real
-#else
-#define RELOC_ENUM int
-#endif
-
#ifndef DEFAULT_ARCH
#define DEFAULT_ARCH "i386"
#endif
static void set_16bit_gcc_code_flag PARAMS ((int));
static void set_intel_syntax PARAMS ((int));
static void set_cpu_arch PARAMS ((int));
+#ifdef TE_PE
+static void pe_directive_secrel PARAMS ((int));
+#endif
static char *output_invalid PARAMS ((int c));
static int i386_operand PARAMS ((char *operand_string));
static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
#define Operand_PCrel 1
/* Relocation type for operand */
- RELOC_ENUM reloc[MAX_OPERANDS];
+ enum bfd_reloc_code_real reloc[MAX_OPERANDS];
/* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
the base index byte below. */
/* List of chars besides those in app.c:symbol_chars that can start an
operand. Used to prevent the scrubber eating vital white-space. */
+const char extra_symbol_chars[] = "*%-(["
#ifdef LEX_AT
-const char extra_symbol_chars[] = "*%-(@[";
-#else
-const char extra_symbol_chars[] = "*%-([";
+ "@"
#endif
+#ifdef LEX_QM
+ "?"
+#endif
+ ;
#if (defined (TE_I386AIX) \
|| ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
&& !defined (TE_LINUX) \
+ && !defined (TE_NETWARE) \
&& !defined (TE_FreeBSD) \
&& !defined (TE_NetBSD)))
/* This array holds the chars that always start a comment. If the
#define is_identifier_char(x) (identifier_chars[(unsigned char) x])
#define is_digit_char(x) (digit_chars[(unsigned char) x])
-/* All non-digit non-letter charcters that may occur in an operand. */
+/* All non-digit non-letter characters that may occur in an operand. */
static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
/* md_assemble() always leaves the strings it's passed unaltered. To
/* CPU name. */
static const char *cpu_arch_name = NULL;
+static const char *cpu_sub_arch_name = NULL;
/* CPU feature flags. */
static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
{"i286", Cpu086|Cpu186|Cpu286 },
{"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
{"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
- {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
- {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
- {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
- {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
- {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
- {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
- {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
- {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
+ {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
+ {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
+ {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
+ {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
+ {"pentiumii", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX },
+ {"pentiumiii",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE },
+ {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
+ {"prescott", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuPNI },
+ {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX },
+ {"k6_2", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
+ {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
+ {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 },
+ {".mmx", CpuMMX },
+ {".sse", CpuMMX|CpuMMX2|CpuSSE },
+ {".sse2", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
+ {".3dnow", CpuMMX|Cpu3dnow },
+ {".3dnowa", CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
+ {".padlock", CpuPadLock },
{NULL, 0 }
};
{"att_syntax", set_intel_syntax, 0},
{"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
{"loc", dwarf2_directive_loc, 0},
+#ifdef TE_PE
+ {"secrel32", pe_directive_secrel, 0},
+#endif
{0, 0, 0}
};
flag_code = new_code_flag;
cpu_arch_flags &= ~(Cpu64 | CpuNo64);
cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
- stackop_size = 'l';
+ stackop_size = LONG_MNEM_SUFFIX;
}
static void
intel_syntax = syntax_flag;
if (ask_naked_reg == 0)
- {
-#ifdef BFD_ASSEMBLER
- allow_naked_reg = (intel_syntax
- && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
-#else
- /* Conservative default. */
- allow_naked_reg = 0;
-#endif
- }
+ allow_naked_reg = (intel_syntax
+ && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
else
allow_naked_reg = (ask_naked_reg < 0);
+
+ identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
+ identifier_chars['$'] = intel_syntax ? '$' : 0;
}
static void
{
if (strcmp (string, cpu_arch[i].name) == 0)
{
- cpu_arch_name = cpu_arch[i].name;
- cpu_arch_flags = (cpu_arch[i].flags
- | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
- break;
+ if (*string != '.')
+ {
+ cpu_arch_name = cpu_arch[i].name;
+ cpu_sub_arch_name = NULL;
+ cpu_arch_flags = (cpu_arch[i].flags
+ | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
+ break;
+ }
+ if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
+ {
+ cpu_sub_arch_name = cpu_arch[i].name;
+ cpu_arch_flags |= cpu_arch[i].flags;
+ }
+ *input_line_pointer = e;
+ demand_empty_rest_of_line ();
+ return;
}
}
if (!cpu_arch[i].name)
demand_empty_rest_of_line ();
}
-#ifdef BFD_ASSEMBLER
unsigned long
i386_mach ()
{
else
as_fatal (_("Unknown architecture"));
}
-#endif
\f
void
md_begin ()
#ifdef LEX_AT
identifier_chars['@'] = '@';
+#endif
+#ifdef LEX_QM
+ identifier_chars['?'] = '?';
+ operand_chars['?'] = '?';
#endif
digit_chars['-'] = '-';
identifier_chars['_'] = '_';
#endif /* DEBUG386 */
\f
-#ifdef BFD_ASSEMBLER
static bfd_reloc_code_real_type reloc
PARAMS ((int, int, int, bfd_reloc_code_real_type));
&& fixP->fx_pcrel)
return 0;
+ /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
+ and changed later by validate_fix. */
+ if (GOT_symbol && fixP->fx_subsy == GOT_symbol
+ && fixP->fx_r_type == BFD_RELOC_32_PCREL)
+ return 0;
+
/* adjust_reloc_syms doesn't know about the GOT. */
if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
|| fixP->fx_r_type == BFD_RELOC_386_PLT32
#endif
return 1;
}
-#else
-#define reloc(SIZE,PCREL,SIGN,OTHER) 0
-#define BFD_RELOC_8 0
-#define BFD_RELOC_16 0
-#define BFD_RELOC_32 0
-#define BFD_RELOC_8_PCREL 0
-#define BFD_RELOC_16_PCREL 0
-#define BFD_RELOC_32_PCREL 0
-#define BFD_RELOC_386_PLT32 0
-#define BFD_RELOC_386_GOT32 0
-#define BFD_RELOC_386_GOTOFF 0
-#define BFD_RELOC_386_TLS_GD 0
-#define BFD_RELOC_386_TLS_LDM 0
-#define BFD_RELOC_386_TLS_LDO_32 0
-#define BFD_RELOC_386_TLS_IE_32 0
-#define BFD_RELOC_386_TLS_IE 0
-#define BFD_RELOC_386_TLS_GOTIE 0
-#define BFD_RELOC_386_TLS_LE_32 0
-#define BFD_RELOC_386_TLS_LE 0
-#define BFD_RELOC_X86_64_PLT32 0
-#define BFD_RELOC_X86_64_GOT32 0
-#define BFD_RELOC_X86_64_GOTPCREL 0
-#define BFD_RELOC_X86_64_TLSGD 0
-#define BFD_RELOC_X86_64_TLSLD 0
-#define BFD_RELOC_X86_64_DTPOFF32 0
-#define BFD_RELOC_X86_64_GOTTPOFF 0
-#define BFD_RELOC_X86_64_TPOFF32 0
-#endif
static int intel_float_operand PARAMS ((const char *mnemonic));
intel_float_operand (mnemonic)
const char *mnemonic;
{
- if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
- return 2;
-
- if (mnemonic[0] == 'f')
- return 1;
+ /* Note that the value returned is meaningful only for opcodes with (memory)
+ operands, hence the code here is free to improperly handle opcodes that
+ have no operands (for better performance and smaller code). */
+
+ if (mnemonic[0] != 'f')
+ return 0; /* non-math */
+
+ switch (mnemonic[1])
+ {
+ /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
+ the fs segment override prefix not currently handled because no
+ call path can make opcodes without operands get here */
+ case 'i':
+ return 2 /* integer op */;
+ case 'l':
+ if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
+ return 3; /* fldcw/fldenv */
+ break;
+ case 'n':
+ if (mnemonic[2] != 'o' /* fnop */)
+ return 3; /* non-waiting control op */
+ break;
+ case 'r':
+ if (mnemonic[2] == 's')
+ return 3; /* frstor/frstpm */
+ break;
+ case 's':
+ if (mnemonic[2] == 'a')
+ return 3; /* fsave */
+ if (mnemonic[2] == 't')
+ {
+ switch (mnemonic[3])
+ {
+ case 'c': /* fstcw */
+ case 'd': /* fstdw */
+ case 'e': /* fstenv */
+ case 's': /* fsts[gw] */
+ return 3;
+ }
+ }
+ break;
+ case 'x':
+ if (mnemonic[2] == 'r' || mnemonic[2] == 's')
+ return 0; /* fxsave/fxrstor are not really math ops */
+ break;
+ }
- return 0;
+ return 1;
}
/* This is the guts of the machine-dependent assembler. LINE points to a
"word ptr" or "byte ptr" on the source operand, but we'll use
the suffix later to choose the destination register. */
if ((i.tm.base_opcode & ~9) == 0x0fb6)
- i.suffix = 0;
+ {
+ if (i.reg_operands < 2
+ && !i.suffix
+ && (~i.tm.opcode_modifier
+ & (No_bSuf
+ | No_wSuf
+ | No_lSuf
+ | No_sSuf
+ | No_xSuf
+ | No_qSuf)))
+ as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
+
+ i.suffix = 0;
+ }
}
if (i.tm.opcode_modifier & FWait)
if (i.tm.opcode_modifier & ImmExt)
{
+ expressionS *exp;
+
+ if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
+ {
+ /* These Intel Prescott New Instructions have the fixed
+ operands with an opcode suffix which is coded in the same
+ place as an 8-bit immediate field would be. Here we check
+ those operands and remove them afterwards. */
+ unsigned int x;
+
+ for (x = 0; x < i.operands; x++)
+ if (i.op[x].regs->reg_num != x)
+ as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
+ i.op[x].regs->reg_name, x + 1, i.tm.name);
+ i.operands = 0;
+ }
+
/* These AMD 3DNow! and Intel Katmai New Instructions have an
opcode suffix which is coded in the same place as an 8-bit
immediate field would be. Here we fake an 8-bit immediate
operand from the opcode suffix stored in tm.extension_opcode. */
- expressionS *exp;
-
assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
exp = &im_expressions[i.imm_operands++];
{
/* In case it is "hi" register, give up. */
if (i.op[x].regs->reg_num > 3)
- as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
+ as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
i.op[x].regs->reg_name);
/* Otherwise it is equivalent to the extended register.
char *l = line;
char *token_start = l;
char *mnem_p;
+ int supported;
+ const template *t;
/* Non-zero if we found a prefix only acceptable with string insns. */
const char *expecting_string_instruction = NULL;
switch (mnem_p[-1])
{
case WORD_MNEM_SUFFIX:
+ if (intel_syntax && (intel_float_operand (mnemonic) & 2))
+ i.suffix = SHORT_MNEM_SUFFIX;
+ else
case BYTE_MNEM_SUFFIX:
case QWORD_MNEM_SUFFIX:
i.suffix = mnem_p[-1];
case 'd':
if (intel_syntax)
{
- if (intel_float_operand (mnemonic))
+ if (intel_float_operand (mnemonic) == 1)
i.suffix = SHORT_MNEM_SUFFIX;
else
i.suffix = LONG_MNEM_SUFFIX;
}
/* Check if instruction is supported on specified architecture. */
- if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
- & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
+ supported = 0;
+ for (t = current_templates->start; t < current_templates->end; ++t)
+ {
+ if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
+ & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
+ supported |= 1;
+ if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
+ supported |= 2;
+ }
+ if (!(supported & 2))
+ {
+ as_bad (flag_code == CODE_64BIT
+ ? _("`%s' is not supported in 64-bit mode")
+ : _("`%s' is only supported in 64-bit mode"),
+ current_templates->start->name);
+ return NULL;
+ }
+ if (!(supported & 1))
{
- as_warn (_("`%s' is not supported on `%s'"),
- current_templates->start->name, cpu_arch_name);
+ as_warn (_("`%s' is not supported on `%s%s'"),
+ current_templates->start->name,
+ cpu_arch_name,
+ cpu_sub_arch_name ? cpu_sub_arch_name : "");
}
else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
{
{
union i386_op temp_op;
unsigned int temp_type;
- RELOC_ENUM temp_reloc;
+ enum bfd_reloc_code_real temp_reloc;
int xchg1 = 0;
int xchg2 = 0;
i.types[op] = Imm64 | Imm32S;
break;
case LONG_MNEM_SUFFIX:
- i.types[op] = Imm32 | Imm64;
+ i.types[op] = Imm32;
break;
case WORD_MNEM_SUFFIX:
- i.types[op] = Imm16 | Imm32 | Imm64;
- break;
+ i.types[op] = Imm16;
break;
case BYTE_MNEM_SUFFIX:
- i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
- break;
+ i.types[op] = Imm8 | Imm8S;
break;
}
break;
: (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
? No_xSuf : 0))))));
- for (t = current_templates->start;
- t < current_templates->end;
- t++)
+ t = current_templates->start;
+ if (i.suffix == QWORD_MNEM_SUFFIX
+ && flag_code != CODE_64BIT
+ && (intel_syntax
+ ? !(t->opcode_modifier & IgnoreSize)
+ && !intel_float_operand (t->name)
+ : intel_float_operand (t->name) != 2)
+ && (!(t->operand_types[0] & (RegMMX | RegXMM))
+ || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM)))
+ && (t->base_opcode != 0x0fc7
+ || t->extension_opcode != 1 /* cmpxchg8b */))
+ t = current_templates->end;
+ for (; t < current_templates->end; t++)
{
/* Must have right number of operands. */
if (i.operands != t->operands)
/* Check the suffix, except for some instructions in intel mode. */
if ((t->opcode_modifier & suffix_check)
&& !(intel_syntax
- && (t->opcode_modifier & IgnoreSize))
- && !(intel_syntax
- && t->base_opcode == 0xd9
- && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
- || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
+ && (t->opcode_modifier & IgnoreSize)))
continue;
/* Do not verify operands when there are none. */
}
static int
-process_suffix ()
+process_suffix (void)
{
/* If matched instruction specifies an explicit instruction mnemonic
suffix, use it. */
Destination register type is more significant than source
register type. */
int op;
+
for (op = i.operands; --op >= 0;)
if ((i.types[op] & Reg)
&& !(i.tm.operand_types[op] & InOutPortReg))
else
abort ();
}
- else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
+ else if ((i.tm.opcode_modifier & DefaultSize)
+ && !i.suffix
+ /* exclude fldenv/frstor/fsave/fstenv */
+ && (i.tm.opcode_modifier & No_sSuf))
{
i.suffix = stackop_size;
}
+ else if (intel_syntax
+ && !i.suffix
+ && ((i.tm.operand_types[0] & JumpAbsolute)
+ || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
+ || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
+ && i.tm.extension_opcode <= 3)))
+ {
+ switch (flag_code)
+ {
+ case CODE_64BIT:
+ if (!(i.tm.opcode_modifier & No_qSuf))
+ {
+ i.suffix = QWORD_MNEM_SUFFIX;
+ break;
+ }
+ case CODE_32BIT:
+ if (!(i.tm.opcode_modifier & No_lSuf))
+ i.suffix = LONG_MNEM_SUFFIX;
+ break;
+ case CODE_16BIT:
+ if (!(i.tm.opcode_modifier & No_wSuf))
+ i.suffix = WORD_MNEM_SUFFIX;
+ break;
+ }
+ }
- /* Change the opcode based on the operand size given by i.suffix;
- We need not change things for byte insns. */
-
- if (!i.suffix && (i.tm.opcode_modifier & W))
+ if (!i.suffix)
{
- as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
- return 0;
+ if (!intel_syntax)
+ {
+ if (i.tm.opcode_modifier & W)
+ {
+ as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
+ return 0;
+ }
+ }
+ else
+ {
+ unsigned int suffixes = ~i.tm.opcode_modifier
+ & (No_bSuf
+ | No_wSuf
+ | No_lSuf
+ | No_sSuf
+ | No_xSuf
+ | No_qSuf);
+
+ if ((i.tm.opcode_modifier & W)
+ || ((suffixes & (suffixes - 1))
+ && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
+ {
+ as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
+ return 0;
+ }
+ }
}
+ /* Change the opcode based on the operand size given by i.suffix;
+ We don't need to change things for byte insns. */
+
if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
{
/* It's not a byte, select word/dword operation. */
size prefix, except for instructions that will ignore this
prefix anyway. */
if (i.suffix != QWORD_MNEM_SUFFIX
- && !(i.tm.opcode_modifier & IgnoreSize)
+ && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
+ && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
&& ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
|| (flag_code == CODE_64BIT
&& (i.tm.opcode_modifier & JumpByte))))
{
unsigned int prefix = DATA_PREFIX_OPCODE;
+
if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
prefix = ADDR_PREFIX_OPCODE;
/* Size floating point instruction. */
if (i.suffix == LONG_MNEM_SUFFIX)
- {
- if (i.tm.opcode_modifier & FloatMF)
- i.tm.base_opcode ^= 4;
- }
+ if (i.tm.opcode_modifier & FloatMF)
+ i.tm.base_opcode ^= 4;
}
return 1;
}
static int
-check_byte_reg ()
+check_byte_reg (void)
{
int op;
+
for (op = i.operands; --op >= 0;)
{
/* If this is an eight bit register, it's OK. If it's the 16 or
|| i.tm.base_opcode == 0xfbf))
continue;
- if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
-#if 0
- /* Check that the template allows eight bit regs. This
- kills insns such as `orb $1,%edx', which maybe should be
- allowed. */
- && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
-#endif
- )
+ if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
{
/* Prohibit these changes in the 64bit mode, since the
lowering is more complicated. */
unsigned int overlap0, overlap1, overlap2;
overlap0 = i.types[0] & i.tm.operand_types[0];
- if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
+ if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
&& overlap0 != Imm8 && overlap0 != Imm8S
&& overlap0 != Imm16 && overlap0 != Imm32S
&& overlap0 != Imm32 && overlap0 != Imm64)
i.types[0] = overlap0;
overlap1 = i.types[1] & i.tm.operand_types[1];
- if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
+ if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
&& overlap1 != Imm8 && overlap1 != Imm8S
&& overlap1 != Imm16 && overlap1 != Imm32S
&& overlap1 != Imm32 && overlap1 != Imm64)
if ((i.op[source].regs->reg_flags & RegRex) != 0)
i.rex |= REX_EXTX;
}
+ if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
+ {
+ if (!((i.types[0] | i.types[1]) & Control))
+ abort ();
+ i.rex &= ~(REX_EXTX | REX_EXTZ);
+ add_prefix (LOCK_PREFIX_OPCODE);
+ }
}
else
{ /* If it's not 2 reg operands... */
if (i.index_reg == 0)
{
/* Operand is just <disp> */
- if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
- && (flag_code != CODE_64BIT))
- {
- i.rm.regmem = NO_BASE_REGISTER_16;
- i.types[op] &= ~Disp;
- i.types[op] |= Disp16;
- }
- else if (flag_code != CODE_64BIT
- || (i.prefix[ADDR_PREFIX] != 0))
- {
- i.rm.regmem = NO_BASE_REGISTER;
- i.types[op] &= ~Disp;
- i.types[op] |= Disp32;
- }
- else
+ if (flag_code == CODE_64BIT)
{
/* 64bit mode overwrites the 32bit absolute
addressing by RIP relative addressing and
i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
i.sib.base = NO_BASE_REGISTER;
i.sib.index = NO_INDEX_REGISTER;
- i.types[op] &= ~Disp;
- i.types[op] |= Disp32S;
+ i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
+ }
+ else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
+ {
+ i.rm.regmem = NO_BASE_REGISTER_16;
+ i.types[op] = Disp16;
+ }
+ else
+ {
+ i.rm.regmem = NO_BASE_REGISTER;
+ i.types[op] = Disp32;
}
}
else /* !i.base_reg && i.index_reg */
else if (i.base_reg->reg_type == BaseIndex)
{
i.rm.regmem = NO_BASE_REGISTER;
- i.types[op] &= ~Disp;
+ i.types[op] &= ~ Disp;
i.types[op] |= Disp32S;
i.flags[op] = Operand_PCrel;
+ if (! i.disp_operands)
+ fake_zero_displacement = 1;
}
else if (i.base_reg->reg_type & Reg16)
{
{
if (flag_code == CODE_64BIT
&& (i.types[op] & Disp))
- {
- if (i.types[op] & Disp8)
- i.types[op] = Disp8 | Disp32S;
- else
- i.types[op] = Disp32S;
- }
+ i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
+
i.rm.regmem = i.base_reg->reg_num;
if ((i.base_reg->reg_flags & RegRex) != 0)
i.rex |= REX_EXTZ;
md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
}
-
static void
output_insn ()
{
char *p;
unsigned char *q;
- /* All opcodes on i386 have either 1 or 2 bytes. We may use third
- byte for the SSE instructions to specify a prefix they require. */
- if (i.tm.base_opcode & 0xff0000)
- add_prefix ((i.tm.base_opcode >> 16) & 0xff);
+ /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
+ have 3 bytes. We may use one more higher byte to specify a prefix
+ the instruction requires. */
+ if ((i.tm.cpu_flags & CpuPadLock) != 0
+ && (i.tm.base_opcode & 0xff000000) != 0)
+ {
+ unsigned int prefix;
+ prefix = (i.tm.base_opcode >> 24) & 0xff;
+
+ if (prefix != REPE_PREFIX_OPCODE
+ || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
+ add_prefix (prefix);
+ }
+ else
+ if ((i.tm.cpu_flags & CpuPadLock) == 0
+ && (i.tm.base_opcode & 0xff0000) != 0)
+ add_prefix ((i.tm.base_opcode >> 16) & 0xff);
/* The prefix bytes. */
for (q = i.prefix;
}
else
{
- p = frag_more (2);
+ if ((i.tm.cpu_flags & CpuPadLock) != 0)
+ {
+ p = frag_more (3);
+ *p++ = (i.tm.base_opcode >> 16) & 0xff;
+ }
+ else
+ p = frag_more (2);
+
/* Put out high byte first: can't use md_number_to_chars! */
*p++ = (i.tm.base_opcode >> 8) & 0xff;
*p = i.tm.base_opcode & 0xff;
}
else
{
- RELOC_ENUM reloc_type;
+ enum bfd_reloc_code_real reloc_type;
int size = 4;
int sign = 0;
int pcrel = (i.flags[n] & Operand_PCrel) != 0;
p = frag_more (size);
reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
-#ifdef BFD_ASSEMBLER
if (reloc_type == BFD_RELOC_32
&& GOT_symbol
&& GOT_symbol == i.op[n].disps->X_add_symbol
reloc_type = BFD_RELOC_386_GOTPC;
i.op[n].disps->X_add_number += add;
}
-#endif
fix_new_exp (frag_now, p - frag_now->fr_literal, size,
i.op[n].disps, pcrel, reloc_type);
}
Need a 32-bit fixup (don't support 8bit
non-absolute imms). Try to support other
sizes ... */
- RELOC_ENUM reloc_type;
+ enum bfd_reloc_code_real reloc_type;
int size = 4;
int sign = 0;
p = frag_more (size);
reloc_type = reloc (size, 0, sign, i.reloc[n]);
-#ifdef BFD_ASSEMBLER
+
/* This is tough to explain. We end up with this one if we
* have operands that look like
* "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
reloc_type = BFD_RELOC_386_GOTPC;
i.op[n].imms->X_add_number += add;
}
-#endif
fix_new_exp (frag_now, p - frag_now->fr_literal, size,
i.op[n].imms, 0, reloc_type);
}
}
\f
#ifndef LEX_AT
-static char *lex_got PARAMS ((RELOC_ENUM *, int *));
+static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
/* Parse operands of the form
<symbol>@GOTOFF+<nnn>
input line. Otherwise return NULL. */
static char *
lex_got (reloc, adjust)
- RELOC_ENUM *reloc;
+ enum bfd_reloc_code_real *reloc;
int *adjust;
{
static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
static const struct {
const char *str;
- const RELOC_ENUM rel[NUM_FLAG_CODE];
+ const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
} gotrel[] = {
{ "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
{ "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
/* x86_cons_fix_new is called via the expression parsing code when a
reloc is needed. We use this hook to get the correct .got reloc. */
-static RELOC_ENUM got_reloc = NO_RELOC;
+static enum bfd_reloc_code_real got_reloc = NO_RELOC;
void
x86_cons_fix_new (frag, off, len, exp)
unsigned int len;
expressionS *exp;
{
- RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
+ enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
got_reloc = NO_RELOC;
fix_new_exp (frag, off, len, exp, 0, r);
}
}
#endif
+#ifdef TE_PE
+
+void
+x86_pe_cons_fix_new (frag, off, len, exp)
+ fragS *frag;
+ unsigned int off;
+ unsigned int len;
+ expressionS *exp;
+{
+ enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
+
+ if (exp->X_op == O_secrel)
+ {
+ exp->X_op = O_symbol;
+ r = BFD_RELOC_32_SECREL;
+ }
+
+ fix_new_exp (frag, off, len, exp, 0, r);
+}
+
+static void
+pe_directive_secrel (dummy)
+ int dummy ATTRIBUTE_UNUSED;
+{
+ expressionS exp;
+
+ do
+ {
+ expression (&exp);
+ if (exp.X_op == O_symbol)
+ exp.X_op = O_secrel;
+
+ emit_expr (&exp, 4);
+ }
+ while (*input_line_pointer++ == ',');
+
+ input_line_pointer--;
+ demand_empty_rest_of_line ();
+}
+
+#endif
+
static int i386_immediate PARAMS ((char *));
static int
exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
}
#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
- else if (1
-#ifdef BFD_ASSEMBLER
- && OUTPUT_FLAVOR == bfd_target_aout_flavour
-#endif
+ else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
&& exp_seg != absolute_section
&& exp_seg != text_section
&& exp_seg != data_section
&& exp_seg != bss_section
&& exp_seg != undefined_section
-#ifdef BFD_ASSEMBLER
- && !bfd_is_com_section (exp_seg)
-#endif
- )
+ && !bfd_is_com_section (exp_seg))
{
-#ifdef BFD_ASSEMBLER
as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
-#else
- as_bad (_("unimplemented segment type %d in operand"), exp_seg);
-#endif
return 0;
}
#endif
switch (val)
{
- case 0:
case 1:
i.log2_scale_factor = 0;
break;
i.log2_scale_factor = 3;
break;
default:
- as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
- scale);
- input_line_pointer = save;
- return NULL;
+ {
+ char sep = *input_line_pointer;
+
+ *input_line_pointer = '\0';
+ as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
+ scale);
+ *input_line_pointer = sep;
+ input_line_pointer = save;
+ return NULL;
+ }
}
if (i.log2_scale_factor != 0 && i.index_reg == 0)
{
free (gotfree_input_line);
#endif
-#ifdef BFD_ASSEMBLER
/* We do this to make sure that the section symbol is in
the symbol table. We will ultimately change the relocation
to be relative to the beginning of the section. */
else
i.reloc[this_operand] = BFD_RELOC_32;
}
-#endif
if (exp->X_op == O_absent || exp->X_op == O_big)
{
#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
if (exp->X_op != O_constant
-#ifdef BFD_ASSEMBLER
&& OUTPUT_FLAVOR == bfd_target_aout_flavour
-#endif
&& exp_seg != absolute_section
&& exp_seg != text_section
&& exp_seg != data_section
&& exp_seg != bss_section
&& exp_seg != undefined_section
-#ifdef BFD_ASSEMBLER
- && !bfd_is_com_section (exp_seg)
-#endif
- )
+ && !bfd_is_com_section (exp_seg))
{
-#ifdef BFD_ASSEMBLER
as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
-#else
- as_bad (_("unimplemented segment type %d in operand"), exp_seg);
-#endif
return 0;
}
#endif
tryprefix:
#endif
ok = 1;
- if (flag_code == CODE_64BIT)
- {
- if (i.prefix[ADDR_PREFIX] == 0)
- {
- /* 64bit checks. */
- if ((i.base_reg
- && ((i.base_reg->reg_type & Reg64) == 0)
- && (i.base_reg->reg_type != BaseIndex
- || i.index_reg))
- || (i.index_reg
- && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
- != (Reg64 | BaseIndex))))
- ok = 0;
- }
- else
- {
- /* 32bit checks. */
- if ((i.base_reg
- && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
- || (i.index_reg
- && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
- != (Reg32 | BaseIndex))))
- ok = 0;
- }
+ if (flag_code == CODE_64BIT)
+ {
+ unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
+
+ if ((i.base_reg
+ && ((i.base_reg->reg_type & RegXX) == 0)
+ && (i.base_reg->reg_type != BaseIndex
+ || i.index_reg))
+ || (i.index_reg
+ && ((i.index_reg->reg_type & (RegXX | BaseIndex))
+ != (RegXX | BaseIndex))))
+ ok = 0;
}
else
{
if (!ok)
{
#if INFER_ADDR_PREFIX
- if (flag_code != CODE_64BIT
- && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
+ if (i.prefix[ADDR_PREFIX] == 0)
{
i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
i.prefixes += 1;
FIXME. There doesn't seem to be any real need for separate
Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
Removing them would probably clean up the code quite a lot. */
- if (i.types[this_operand] & (Disp16 | Disp32))
+ if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
i.types[this_operand] ^= (Disp16 | Disp32);
fudged = 1;
goto tryprefix;
as_bad (_("`%s' is not a valid %s bit base/index expression"),
operand_string,
flag_code_names[flag_code]);
- return 0;
}
- return 1;
+ return ok;
}
/* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
/* Symbol is undefined in this segment, or we need to keep a
reloc so that weak symbols can be overridden. */
int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
- RELOC_ENUM reloc_type;
+ enum bfd_reloc_code_real reloc_type;
unsigned char *opcode;
int old_fr_fix;
Out: Any fixSs and constants are set up.
Caller will turn frag into a ".space 0". */
-#ifndef BFD_ASSEMBLER
-void
-md_convert_frag (headers, sec, fragP)
- object_headers *headers ATTRIBUTE_UNUSED;
- segT sec ATTRIBUTE_UNUSED;
- fragS *fragP;
-#else
void
md_convert_frag (abfd, sec, fragP)
bfd *abfd ATTRIBUTE_UNUSED;
segT sec ATTRIBUTE_UNUSED;
fragS *fragP;
-#endif
{
unsigned char *opcode;
unsigned char *where_to_put_displacement = NULL;
char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
valueT value = *valP;
-#if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
+#if !defined (TE_Mach)
if (fixP->fx_pcrel)
{
switch (fixP->fx_r_type)
}
#endif
#if defined (OBJ_COFF) && defined (TE_PE)
- /* For some reason, the PE format does not store a section
- address offset for a PC relative symbol. */
- if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
+ /* For some reason, the PE format does not store a
+ section address offset for a PC relative symbol. */
+ if (S_GET_SEGMENT (fixP->fx_addsy) != seg
+#if defined(BFD_ASSEMBLER) || defined(S_IS_WEAK)
+ || S_IS_WEAK (fixP->fx_addsy)
+#endif
+ )
value += md_pcrel_from (fixP);
#endif
}
/* Fix a few things - the dynamic linker expects certain values here,
- and we must not dissappoint it. */
+ and we must not disappoint it. */
#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
if (OUTPUT_FLAVOR == bfd_target_elf_flavour
&& fixP->fx_addsy)
}
#endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
*valP = value;
-#endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
+#endif /* !defined (TE_Mach) */
/* Are we finished with this relocation now? */
if (fixP->fx_addsy == NULL)
fixP->fx_done = 1;
-#ifdef BFD_ASSEMBLER
else if (use_rela_relocations)
{
fixP->fx_no_overflow = 1;
fixP->fx_addnumber = value;
value = 0;
}
-#endif
+
md_number_to_chars (p, value, fixP->fx_size);
}
\f
}
if (r != NULL
- && (r->reg_flags & (RegRex64 | RegRex)) != 0
+ && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
+ && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
&& flag_code != CODE_64BIT)
- {
- return (const reg_entry *) NULL;
- }
+ return (const reg_entry *) NULL;
return r;
}
#endif
}
-#ifdef BFD_ASSEMBLER
#if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
|| defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
}
}
#endif
-#endif /* BFD_ASSEMBLER */
\f
symbolS *
md_undefined_symbol (name)
segT segment ATTRIBUTE_UNUSED;
valueT size;
{
-#ifdef BFD_ASSEMBLER
#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
{
align = bfd_get_section_alignment (stdoutput, segment);
size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
}
-#endif
#endif
return size;
#endif
-#ifdef BFD_ASSEMBLER
-
void
i386_validate_fix (fixp)
fixS *fixp;
case BFD_RELOC_RVA:
case BFD_RELOC_VTABLE_ENTRY:
case BFD_RELOC_VTABLE_INHERIT:
+#ifdef TE_PE
+ case BFD_RELOC_32_SECREL:
+#endif
code = fixp->fx_r_type;
break;
default:
*rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
+
if (!use_rela_relocations)
{
/* HACK: Since i386 ELF uses Rel instead of Rela, encode the
return rel;
}
-#else /* !BFD_ASSEMBLER */
-
-#if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
-void
-tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
- char *where;
- fixS *fixP;
- relax_addressT segment_address_in_file;
-{
- /* In: length of relocation (or of address) in chars: 1, 2 or 4.
- Out: GNU LD relocation length code: 0, 1, or 2. */
-
- static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
- long r_symbolnum;
-
- know (fixP->fx_addsy != NULL);
-
- md_number_to_chars (where,
- (valueT) (fixP->fx_frag->fr_address
- + fixP->fx_where - segment_address_in_file),
- 4);
-
- r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
- ? S_GET_TYPE (fixP->fx_addsy)
- : fixP->fx_addsy->sy_number);
-
- where[6] = (r_symbolnum >> 16) & 0x0ff;
- where[5] = (r_symbolnum >> 8) & 0x0ff;
- where[4] = r_symbolnum & 0x0ff;
- where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
- | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
- | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
-}
-
-#endif /* OBJ_AOUT or OBJ_BOUT. */
-
-#if defined (I386COFF)
-
-short
-tc_coff_fix2rtype (fixP)
- fixS *fixP;
-{
- if (fixP->fx_r_type == R_IMAGEBASE)
- return R_IMAGEBASE;
-
- return (fixP->fx_pcrel ?
- (fixP->fx_size == 1 ? R_PCRBYTE :
- fixP->fx_size == 2 ? R_PCRWORD :
- R_PCRLONG) :
- (fixP->fx_size == 1 ? R_RELBYTE :
- fixP->fx_size == 2 ? R_RELWORD :
- R_DIR32));
-}
-
-int
-tc_coff_sizemachdep (frag)
- fragS *frag;
-{
- if (frag->fr_next)
- return (frag->fr_next->fr_address - frag->fr_address);
- else
- return 0;
-}
-
-#endif /* I386COFF */
-
-#endif /* !BFD_ASSEMBLER */
\f
/* Parse operands using Intel syntax. This implements a recursive descent
parser based on the BNF grammar published in Appendix B of the MASM 6.1
alpha [a-zA-Z]
+ binOp & | AND | \| | OR | ^ | XOR
+
byteRegister AL | AH | BL | BH | CL | CH | DL | DH
constant digits [[ radixOverride ]]
- dataType BYTE | WORD | DWORD | QWORD | XWORD
+ dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
digits decdigit
| digits decdigit
decdigit [0-9]
- e05 e05 addOp e06
+ e04 e04 addOp e05
+ | e05
+
+ e05 e05 binOp e06
| e06
e06 e06 mulOp e09
| e09
e09 OFFSET e10
+ | SHORT e10
+ | + e10
+ | - e10
+ | ~ e10
+ | NOT e10
| e09 PTR e10
| e09 : e10
| e10
| $
| register
- => expr SHORT e05
- | e05
+ => expr expr cmpOp e04
+ | e04
gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
| BP | EBP | SP | ESP | DI | EDI | SI | ESI
| id alpha
| id decdigit
- mulOp * | / | MOD
+ mulOp * | / | % | MOD | << | SHL | >> | SHR
quote " | '
segmentRegister CS | DS | ES | FS | GS | SS
- specialRegister CR0 | CR2 | CR3
+ specialRegister CR0 | CR2 | CR3 | CR4
| DR0 | DR1 | DR2 | DR3 | DR6 | DR7
| TR3 | TR4 | TR5 | TR6 | TR7
done by calling parse_register) and eliminate immediate left recursion
to implement a recursive-descent parser.
- expr SHORT e05
- | e05
+ expr e04 expr'
+
+ expr' cmpOp e04 expr'
+ | Empty
+
+ e04 e05 e04'
+
+ e04' addOp e05 e04'
+ | Empty
e05 e06 e05'
- e05' addOp e06 e05'
+ e05' binOp e06 e05'
| Empty
e06 e09 e06'
| Empty
e09 OFFSET e10 e09'
+ | SHORT e10'
+ | + e10'
+ | - e10'
+ | ~ e10'
+ | NOT e10'
| e10 e09'
e09' PTR e10 e09'
| BYTE
| WORD
| DWORD
+ | FWORD
| QWORD
- | XWORD
+ | TBYTE
+ | OWORD
+ | XMMWORD
| .
| $
| register
int got_a_float; /* Whether the operand is a float. */
int op_modifier; /* Operand modifier. */
int is_mem; /* 1 if operand is memory reference. */
+ int in_offset; /* >=1 if parsing operand of offset. */
+ int in_bracket; /* >=1 if parsing operand in brackets. */
const reg_entry *reg; /* Last register reference found. */
char *disp; /* Displacement string being built. */
+ char *next_operand; /* Resume point when splitting operands. */
};
static struct intel_parser_s intel_parser;
#define T_REG 2
#define T_BYTE 3
#define T_WORD 4
-#define T_DWORD 5
-#define T_QWORD 6
-#define T_XWORD 7
+#define T_DWORD 5
+#define T_FWORD 6
+#define T_QWORD 7
+#define T_TBYTE 8
+#define T_XMMWORD 9
#undef T_SHORT
-#define T_SHORT 8
-#define T_OFFSET 9
-#define T_PTR 10
-#define T_ID 11
+#define T_SHORT 10
+#define T_OFFSET 11
+#define T_PTR 12
+#define T_ID 13
+#define T_SHL 14
+#define T_SHR 15
/* Prototypes for intel parser functions. */
static int intel_match_token PARAMS ((int code));
static void intel_get_token PARAMS ((void));
static void intel_putback_token PARAMS ((void));
static int intel_expr PARAMS ((void));
+static int intel_e04 PARAMS ((void));
static int intel_e05 PARAMS ((void));
-static int intel_e05_1 PARAMS ((void));
static int intel_e06 PARAMS ((void));
-static int intel_e06_1 PARAMS ((void));
static int intel_e09 PARAMS ((void));
-static int intel_e09_1 PARAMS ((void));
+static int intel_bracket_expr PARAMS ((void));
static int intel_e10 PARAMS ((void));
-static int intel_e10_1 PARAMS ((void));
static int intel_e11 PARAMS ((void));
static int
int ret;
char *p;
- /* Initialize token holders. */
- cur_token.code = prev_token.code = T_NIL;
- cur_token.reg = prev_token.reg = NULL;
- cur_token.str = prev_token.str = NULL;
-
- /* Initialize parser structure. */
- p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
- if (p == NULL)
- abort ();
- strcpy (intel_parser.op_string, operand_string);
- intel_parser.got_a_float = got_a_float;
- intel_parser.op_modifier = -1;
- intel_parser.is_mem = 0;
- intel_parser.reg = NULL;
- intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
- if (intel_parser.disp == NULL)
- abort ();
- intel_parser.disp[0] = '\0';
-
- /* Read the first token and start the parser. */
- intel_get_token ();
- ret = intel_expr ();
-
- if (ret)
+ p = intel_parser.op_string = xstrdup (operand_string);
+ intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
+
+ for (;;)
{
+ /* Initialize token holders. */
+ cur_token.code = prev_token.code = T_NIL;
+ cur_token.reg = prev_token.reg = NULL;
+ cur_token.str = prev_token.str = NULL;
+
+ /* Initialize parser structure. */
+ intel_parser.got_a_float = got_a_float;
+ intel_parser.op_modifier = 0;
+ intel_parser.is_mem = 0;
+ intel_parser.in_offset = 0;
+ intel_parser.in_bracket = 0;
+ intel_parser.reg = NULL;
+ intel_parser.disp[0] = '\0';
+ intel_parser.next_operand = NULL;
+
+ /* Read the first token and start the parser. */
+ intel_get_token ();
+ ret = intel_expr ();
+
+ if (!ret)
+ break;
+
+ if (cur_token.code != T_NIL)
+ {
+ as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
+ current_templates->start->name, cur_token.str);
+ ret = 0;
+ }
/* If we found a memory reference, hand it over to i386_displacement
to fill in the rest of the operand fields. */
- if (intel_parser.is_mem)
+ else if (intel_parser.is_mem)
{
if ((i.mem_operands == 1
&& (current_templates->start->opcode_modifier & IsString) == 0)
char *s = intel_parser.disp;
i.mem_operands++;
+ if (!quiet_warnings && intel_parser.is_mem < 0)
+ /* See the comments in intel_bracket_expr. */
+ as_warn (_("Treating `%s' as memory reference"), operand_string);
+
/* Add the displacement expression. */
if (*s != '\0')
- ret = i386_displacement (s, s + strlen (s))
- && i386_index_check (s);
- }
+ ret = i386_displacement (s, s + strlen (s));
+ if (ret)
+ {
+ /* Swap base and index in 16-bit memory operands like
+ [si+bx]. Since i386_index_check is also used in AT&T
+ mode we have to do that here. */
+ if (i.base_reg
+ && i.index_reg
+ && (i.base_reg->reg_type & Reg16)
+ && (i.index_reg->reg_type & Reg16)
+ && i.base_reg->reg_num >= 6
+ && i.index_reg->reg_num < 6)
+ {
+ const reg_entry *base = i.index_reg;
+
+ i.index_reg = i.base_reg;
+ i.base_reg = base;
+ }
+ ret = i386_index_check (operand_string);
+ }
+ }
}
/* Constant and OFFSET expressions are handled by i386_immediate. */
- else if (intel_parser.op_modifier == OFFSET_FLAT
+ else if ((intel_parser.op_modifier & (1 << T_OFFSET))
|| intel_parser.reg == NULL)
ret = i386_immediate (intel_parser.disp);
+
+ if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
+ ret = 0;
+ if (!ret || !intel_parser.next_operand)
+ break;
+ intel_parser.op_string = intel_parser.next_operand;
+ this_operand = i.operands++;
}
free (p);
return ret;
}
-/* expr SHORT e05
- | e05 */
+#define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
+
+/* expr e04 expr'
+
+ expr' cmpOp e04 expr'
+ | Empty */
static int
intel_expr ()
{
- /* expr SHORT e05 */
- if (cur_token.code == T_SHORT)
+ /* XXX Implement the comparison operators. */
+ return intel_e04 ();
+}
+
+/* e04 e05 e04'
+
+ e04' addOp e05 e04'
+ | Empty */
+static int
+intel_e04 ()
+{
+ int nregs = -1;
+
+ for (;;)
{
- intel_parser.op_modifier = SHORT;
- intel_match_token (T_SHORT);
+ if (!intel_e05())
+ return 0;
- return (intel_e05 ());
- }
+ if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
+ i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
- /* expr e05 */
- else
- return intel_e05 ();
+ if (cur_token.code == '+')
+ nregs = -1;
+ else if (cur_token.code == '-')
+ nregs = NUM_ADDRESS_REGS;
+ else
+ return 1;
+
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+ }
}
/* e05 e06 e05'
- e05' addOp e06 e05'
+ e05' binOp e06 e05'
| Empty */
static int
intel_e05 ()
{
- return (intel_e06 () && intel_e05_1 ());
-}
+ int nregs = ~NUM_ADDRESS_REGS;
-static int
-intel_e05_1 ()
-{
- /* e05' addOp e06 e05' */
- if (cur_token.code == '+' || cur_token.code == '-')
+ for (;;)
{
- strcat (intel_parser.disp, cur_token.str);
+ if (!intel_e06())
+ return 0;
+
+ if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
+ {
+ char str[2];
+
+ str[0] = cur_token.code;
+ str[1] = 0;
+ strcat (intel_parser.disp, str);
+ }
+ else
+ break;
+
intel_match_token (cur_token.code);
- return (intel_e06 () && intel_e05_1 ());
+ if (nregs < 0)
+ nregs = ~nregs;
}
-
- /* e05' Empty */
- else
- return 1;
+ if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
+ i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
+ return 1;
}
/* e06 e09 e06'
static int
intel_e06 ()
{
- return (intel_e09 () && intel_e06_1 ());
-}
+ int nregs = ~NUM_ADDRESS_REGS;
-static int
-intel_e06_1 ()
-{
- /* e06' mulOp e09 e06' */
- if (cur_token.code == '*' || cur_token.code == '/')
+ for (;;)
{
- strcat (intel_parser.disp, cur_token.str);
- intel_match_token (cur_token.code);
+ if (!intel_e09())
+ return 0;
- return (intel_e09 () && intel_e06_1 ());
- }
+ if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
+ {
+ char str[2];
- /* e06' Empty */
- else
- return 1;
+ str[0] = cur_token.code;
+ str[1] = 0;
+ strcat (intel_parser.disp, str);
+ }
+ else if (cur_token.code == T_SHL)
+ strcat (intel_parser.disp, "<<");
+ else if (cur_token.code == T_SHR)
+ strcat (intel_parser.disp, ">>");
+ else
+ break;
+
+ intel_match_token (cur_token.code);
+
+ if (nregs < 0)
+ nregs = ~nregs;
+ }
+ if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
+ i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
+ return 1;
}
-/* e09 OFFSET e10 e09'
+/* e09 OFFSET e09
+ | SHORT e09
+ | + e09
+ | - e09
+ | ~ e09
+ | NOT e09
| e10 e09'
e09' PTR e10 e09'
static int
intel_e09 ()
{
- /* e09 OFFSET e10 e09' */
- if (cur_token.code == T_OFFSET)
+ int nregs = ~NUM_ADDRESS_REGS;
+ int in_offset = 0;
+
+ for (;;)
{
- intel_parser.is_mem = 0;
- intel_parser.op_modifier = OFFSET_FLAT;
- intel_match_token (T_OFFSET);
+ /* Don't consume constants here. */
+ if (cur_token.code == '+' || cur_token.code == '-')
+ {
+ /* Need to look one token ahead - if the next token
+ is a constant, the current token is its sign. */
+ int next_code;
- return (intel_e10 () && intel_e09_1 ());
- }
+ intel_match_token (cur_token.code);
+ next_code = cur_token.code;
+ intel_putback_token ();
+ if (next_code == T_CONST)
+ break;
+ }
- /* e09 e10 e09' */
- else
- return (intel_e10 () && intel_e09_1 ());
-}
+ /* e09 OFFSET e09 */
+ if (cur_token.code == T_OFFSET)
+ {
+ if (!in_offset++)
+ ++intel_parser.in_offset;
+ }
-static int
-intel_e09_1 ()
-{
- /* e09' PTR e10 e09' */
- if (cur_token.code == T_PTR)
- {
- if (prev_token.code == T_BYTE)
- i.suffix = BYTE_MNEM_SUFFIX;
+ /* e09 SHORT e09 */
+ else if (cur_token.code == T_SHORT)
+ intel_parser.op_modifier |= 1 << T_SHORT;
+
+ /* e09 + e09 */
+ else if (cur_token.code == '+')
+ strcat (intel_parser.disp, "+");
- else if (prev_token.code == T_WORD)
+ /* e09 - e09
+ | ~ e09
+ | NOT e09 */
+ else if (cur_token.code == '-' || cur_token.code == '~')
{
- if (intel_parser.got_a_float == 2) /* "fi..." */
- i.suffix = SHORT_MNEM_SUFFIX;
- else
- i.suffix = WORD_MNEM_SUFFIX;
+ char str[2];
+
+ if (nregs < 0)
+ nregs = ~nregs;
+ str[0] = cur_token.code;
+ str[1] = 0;
+ strcat (intel_parser.disp, str);
}
- else if (prev_token.code == T_DWORD)
+ /* e09 e10 e09' */
+ else
+ break;
+
+ intel_match_token (cur_token.code);
+ }
+
+ for (;;)
+ {
+ if (!intel_e10 ())
+ return 0;
+
+ /* e09' PTR e10 e09' */
+ if (cur_token.code == T_PTR)
{
- if (intel_parser.got_a_float == 1) /* "f..." */
- i.suffix = SHORT_MNEM_SUFFIX;
+ char suffix;
+
+ if (prev_token.code == T_BYTE)
+ suffix = BYTE_MNEM_SUFFIX;
+
+ else if (prev_token.code == T_WORD)
+ {
+ if (current_templates->start->name[0] == 'l'
+ && current_templates->start->name[2] == 's'
+ && current_templates->start->name[3] == 0)
+ suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
+ else if (intel_parser.got_a_float == 2) /* "fi..." */
+ suffix = SHORT_MNEM_SUFFIX;
+ else
+ suffix = WORD_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_DWORD)
+ {
+ if (current_templates->start->name[0] == 'l'
+ && current_templates->start->name[2] == 's'
+ && current_templates->start->name[3] == 0)
+ suffix = WORD_MNEM_SUFFIX;
+ else if (flag_code == CODE_16BIT
+ && (current_templates->start->opcode_modifier
+ & (Jump|JumpDword|JumpInterSegment)))
+ suffix = LONG_DOUBLE_MNEM_SUFFIX;
+ else if (intel_parser.got_a_float == 1) /* "f..." */
+ suffix = SHORT_MNEM_SUFFIX;
+ else
+ suffix = LONG_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_FWORD)
+ {
+ if (current_templates->start->name[0] == 'l'
+ && current_templates->start->name[2] == 's'
+ && current_templates->start->name[3] == 0)
+ suffix = LONG_MNEM_SUFFIX;
+ else if (!intel_parser.got_a_float)
+ {
+ if (flag_code == CODE_16BIT)
+ add_prefix (DATA_PREFIX_OPCODE);
+ suffix = LONG_DOUBLE_MNEM_SUFFIX;
+ }
+ else
+ suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
+ }
+
+ else if (prev_token.code == T_QWORD)
+ {
+ if (intel_parser.got_a_float == 1) /* "f..." */
+ suffix = LONG_MNEM_SUFFIX;
+ else
+ suffix = QWORD_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_TBYTE)
+ {
+ if (intel_parser.got_a_float == 1)
+ suffix = LONG_DOUBLE_MNEM_SUFFIX;
+ else
+ suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
+ }
+
+ else if (prev_token.code == T_XMMWORD)
+ {
+ /* XXX ignored for now, but accepted since gcc uses it */
+ suffix = 0;
+ }
+
else
- i.suffix = LONG_MNEM_SUFFIX;
+ {
+ as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
+ return 0;
+ }
+
+ if (current_templates->start->base_opcode == 0x8d /* lea */)
+ ;
+ else if (!i.suffix)
+ i.suffix = suffix;
+ else if (i.suffix != suffix)
+ {
+ as_bad (_("Conflicting operand modifiers"));
+ return 0;
+ }
+
}
- else if (prev_token.code == T_QWORD)
+ /* e09' : e10 e09' */
+ else if (cur_token.code == ':')
{
- if (intel_parser.got_a_float == 1) /* "f..." */
- i.suffix = LONG_MNEM_SUFFIX;
- else
- i.suffix = QWORD_MNEM_SUFFIX;
+ if (prev_token.code != T_REG)
+ {
+ /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
+ segment/group identifier (which we don't have), using comma
+ as the operand separator there is even less consistent, since
+ there all branches only have a single operand. */
+ if (this_operand != 0
+ || intel_parser.in_offset
+ || intel_parser.in_bracket
+ || (!(current_templates->start->opcode_modifier
+ & (Jump|JumpDword|JumpInterSegment))
+ && !(current_templates->start->operand_types[0]
+ & JumpAbsolute)))
+ return intel_match_token (T_NIL);
+ /* Remember the start of the 2nd operand and terminate 1st
+ operand here.
+ XXX This isn't right, yet (when SSSS:OOOO is right operand of
+ another expression), but it gets at least the simplest case
+ (a plain number or symbol on the left side) right. */
+ intel_parser.next_operand = intel_parser.op_string;
+ *--intel_parser.op_string = '\0';
+ return intel_match_token (':');
+ }
}
- else if (prev_token.code == T_XWORD)
- i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
-
+ /* e09' Empty */
else
+ break;
+
+ intel_match_token (cur_token.code);
+
+ }
+
+ if (in_offset)
+ {
+ --intel_parser.in_offset;
+ if (nregs < 0)
+ nregs = ~nregs;
+ if (NUM_ADDRESS_REGS > nregs)
{
- as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
+ as_bad (_("Invalid operand to `OFFSET'"));
return 0;
}
+ intel_parser.op_modifier |= 1 << T_OFFSET;
+ }
- intel_match_token (T_PTR);
+ if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
+ i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
+ return 1;
+}
- return (intel_e10 () && intel_e09_1 ());
+static int
+intel_bracket_expr ()
+{
+ int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
+ const char *start = intel_parser.op_string;
+ int len;
+
+ if (i.op[this_operand].regs)
+ return intel_match_token (T_NIL);
+
+ intel_match_token ('[');
+
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. If it's an offset expression, we need to keep
+ the brace in. */
+ if (!intel_parser.in_offset)
+ {
+ ++intel_parser.in_bracket;
+ /* Unfortunately gas always diverged from MASM in a respect that can't
+ be easily fixed without risking to break code sequences likely to be
+ encountered (the testsuite even check for this): MASM doesn't consider
+ an expression inside brackets unconditionally as a memory reference.
+ When that is e.g. a constant, an offset expression, or the sum of the
+ two, this is still taken as a constant load. gas, however, always
+ treated these as memory references. As a compromise, we'll try to make
+ offset expressions inside brackets work the MASM way (since that's
+ less likely to be found in real world code), but make constants alone
+ continue to work the traditional gas way. In either case, issue a
+ warning. */
+ intel_parser.op_modifier &= ~was_offset;
}
+ else
+ strcat (intel_parser.disp, "[");
+
+ /* Add a '+' to the displacement string if necessary. */
+ if (*intel_parser.disp != '\0'
+ && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
+ strcat (intel_parser.disp, "+");
- /* e09 : e10 e09' */
- else if (cur_token.code == ':')
+ if (intel_expr ()
+ && (len = intel_parser.op_string - start - 1,
+ intel_match_token (']')))
{
- /* Mark as a memory operand only if it's not already known to be an
- offset expression. */
- if (intel_parser.op_modifier != OFFSET_FLAT)
- intel_parser.is_mem = 1;
+ /* Preserve brackets when the operand is an offset expression. */
+ if (intel_parser.in_offset)
+ strcat (intel_parser.disp, "]");
+ else
+ {
+ --intel_parser.in_bracket;
+ if (i.base_reg || i.index_reg)
+ intel_parser.is_mem = 1;
+ if (!intel_parser.is_mem)
+ {
+ if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
+ /* Defer the warning until all of the operand was parsed. */
+ intel_parser.is_mem = -1;
+ else if (!quiet_warnings)
+ as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start);
+ }
+ }
+ intel_parser.op_modifier |= was_offset;
- return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
+ return 1;
}
-
- /* e09' Empty */
- else
- return 1;
+ return 0;
}
/* e10 e11 e10'
static int
intel_e10 ()
{
- return (intel_e11 () && intel_e10_1 ());
-}
+ if (!intel_e11 ())
+ return 0;
-static int
-intel_e10_1 ()
-{
- /* e10' [ expr ] e10' */
- if (cur_token.code == '[')
+ while (cur_token.code == '[')
{
- intel_match_token ('[');
-
- /* Mark as a memory operand only if it's not already known to be an
- offset expression. If it's an offset expression, we need to keep
- the brace in. */
- if (intel_parser.op_modifier != OFFSET_FLAT)
- intel_parser.is_mem = 1;
- else
- strcat (intel_parser.disp, "[");
-
- /* Add a '+' to the displacement string if necessary. */
- if (*intel_parser.disp != '\0'
- && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
- strcat (intel_parser.disp, "+");
-
- if (intel_expr () && intel_match_token (']'))
- {
- /* Preserve brackets when the operand is an offset expression. */
- if (intel_parser.op_modifier == OFFSET_FLAT)
- strcat (intel_parser.disp, "]");
-
- return intel_e10_1 ();
- }
- else
+ if (!intel_bracket_expr ())
return 0;
}
- /* e10' Empty */
- else
- return 1;
+ return 1;
}
/* e11 ( expr )
| BYTE
| WORD
| DWORD
+ | FWORD
| QWORD
- | XWORD
+ | TBYTE
+ | OWORD
+ | XMMWORD
| $
| .
| register
static int
intel_e11 ()
{
- /* e11 ( expr ) */
- if (cur_token.code == '(')
+ switch (cur_token.code)
{
+ /* e11 ( expr ) */
+ case '(':
intel_match_token ('(');
strcat (intel_parser.disp, "(");
strcat (intel_parser.disp, ")");
return 1;
}
- else
- return 0;
- }
-
- /* e11 [ expr ] */
- else if (cur_token.code == '[')
- {
- intel_match_token ('[');
-
- /* Mark as a memory operand only if it's not already known to be an
- offset expression. If it's an offset expression, we need to keep
- the brace in. */
- if (intel_parser.op_modifier != OFFSET_FLAT)
- intel_parser.is_mem = 1;
- else
- strcat (intel_parser.disp, "[");
+ return 0;
- /* Operands for jump/call inside brackets denote absolute addresses. */
- if (current_templates->start->opcode_modifier & Jump
- || current_templates->start->opcode_modifier & JumpDword
- || current_templates->start->opcode_modifier & JumpByte
- || current_templates->start->opcode_modifier & JumpInterSegment)
+ /* e11 [ expr ] */
+ case '[':
+ /* Operands for jump/call inside brackets denote absolute addresses.
+ XXX This shouldn't be needed anymore (or if it should rather live
+ in intel_bracket_expr). */
+ if (current_templates->start->opcode_modifier
+ & (Jump|JumpDword|JumpByte|JumpInterSegment))
i.types[this_operand] |= JumpAbsolute;
- /* Add a '+' to the displacement string if necessary. */
- if (*intel_parser.disp != '\0'
- && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
- strcat (intel_parser.disp, "+");
-
- if (intel_expr () && intel_match_token (']'))
- {
- /* Preserve brackets when the operand is an offset expression. */
- if (intel_parser.op_modifier == OFFSET_FLAT)
- strcat (intel_parser.disp, "]");
-
- return 1;
- }
- else
- return 0;
- }
+ return intel_bracket_expr ();
- /* e11 BYTE
- | WORD
- | DWORD
- | QWORD
- | XWORD */
- else if (cur_token.code == T_BYTE
- || cur_token.code == T_WORD
- || cur_token.code == T_DWORD
- || cur_token.code == T_QWORD
- || cur_token.code == T_XWORD)
- {
- intel_match_token (cur_token.code);
-
- return 1;
- }
-
- /* e11 $
- | . */
- else if (cur_token.code == '$' || cur_token.code == '.')
- {
+ /* e11 $
+ | . */
+ case '.':
strcat (intel_parser.disp, cur_token.str);
intel_match_token (cur_token.code);
/* Mark as a memory operand only if it's not already known to be an
offset expression. */
- if (intel_parser.op_modifier != OFFSET_FLAT)
+ if (!intel_parser.in_offset)
intel_parser.is_mem = 1;
return 1;
- }
- /* e11 register */
- else if (cur_token.code == T_REG)
- {
- const reg_entry *reg = intel_parser.reg = cur_token.reg;
+ /* e11 register */
+ case T_REG:
+ {
+ const reg_entry *reg = intel_parser.reg = cur_token.reg;
- intel_match_token (T_REG);
+ intel_match_token (T_REG);
- /* Check for segment change. */
- if (cur_token.code == ':')
- {
- if (reg->reg_type & (SReg2 | SReg3))
- {
- switch (reg->reg_num)
- {
- case 0:
- i.seg[i.mem_operands] = &es;
- break;
- case 1:
- i.seg[i.mem_operands] = &cs;
- break;
- case 2:
- i.seg[i.mem_operands] = &ss;
- break;
- case 3:
- i.seg[i.mem_operands] = &ds;
- break;
- case 4:
- i.seg[i.mem_operands] = &fs;
- break;
- case 5:
- i.seg[i.mem_operands] = &gs;
- break;
- }
- }
- else
- {
- as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
- return 0;
- }
- }
+ /* Check for segment change. */
+ if (cur_token.code == ':')
+ {
+ if (!(reg->reg_type & (SReg2 | SReg3)))
+ {
+ as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
+ return 0;
+ }
+ else if (i.seg[i.mem_operands])
+ as_warn (_("Extra segment override ignored"));
+ else
+ {
+ if (!intel_parser.in_offset)
+ intel_parser.is_mem = 1;
+ switch (reg->reg_num)
+ {
+ case 0:
+ i.seg[i.mem_operands] = &es;
+ break;
+ case 1:
+ i.seg[i.mem_operands] = &cs;
+ break;
+ case 2:
+ i.seg[i.mem_operands] = &ss;
+ break;
+ case 3:
+ i.seg[i.mem_operands] = &ds;
+ break;
+ case 4:
+ i.seg[i.mem_operands] = &fs;
+ break;
+ case 5:
+ i.seg[i.mem_operands] = &gs;
+ break;
+ }
+ }
+ }
- /* Not a segment register. Check for register scaling. */
- else if (cur_token.code == '*')
- {
- if (!intel_parser.is_mem)
- {
- as_bad (_("Register scaling only allowed in memory operands."));
- return 0;
- }
+ /* Not a segment register. Check for register scaling. */
+ else if (cur_token.code == '*')
+ {
+ if (!intel_parser.in_bracket)
+ {
+ as_bad (_("Register scaling only allowed in memory operands"));
+ return 0;
+ }
- /* What follows must be a valid scale. */
- if (intel_match_token ('*')
- && strchr ("01248", *cur_token.str))
- {
- i.index_reg = reg;
- i.types[this_operand] |= BaseIndex;
+ if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
+ reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
+ else if (i.index_reg)
+ reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
- /* Set the scale after setting the register (otherwise,
- i386_scale will complain) */
- i386_scale (cur_token.str);
- intel_match_token (T_CONST);
- }
- else
- {
- as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
- cur_token.str);
- return 0;
- }
- }
+ /* What follows must be a valid scale. */
+ intel_match_token ('*');
+ i.index_reg = reg;
+ i.types[this_operand] |= BaseIndex;
- /* No scaling. If this is a memory operand, the register is either a
- base register (first occurrence) or an index register (second
- occurrence). */
- else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
- {
- if (i.base_reg && i.index_reg)
- {
- as_bad (_("Too many register references in memory operand.\n"));
+ /* Set the scale after setting the register (otherwise,
+ i386_scale will complain) */
+ if (cur_token.code == '+' || cur_token.code == '-')
+ {
+ char *str, sign = cur_token.code;
+ intel_match_token (cur_token.code);
+ if (cur_token.code != T_CONST)
+ {
+ as_bad (_("Syntax error: Expecting a constant, got `%s'"),
+ cur_token.str);
+ return 0;
+ }
+ str = (char *) xmalloc (strlen (cur_token.str) + 2);
+ strcpy (str + 1, cur_token.str);
+ *str = sign;
+ if (!i386_scale (str))
+ return 0;
+ free (str);
+ }
+ else if (!i386_scale (cur_token.str))
return 0;
- }
+ intel_match_token (cur_token.code);
+ }
- if (i.base_reg == NULL)
- i.base_reg = reg;
- else
- i.index_reg = reg;
+ /* No scaling. If this is a memory operand, the register is either a
+ base register (first occurrence) or an index register (second
+ occurrence). */
+ else if (intel_parser.in_bracket && !(reg->reg_type & (SReg2 | SReg3)))
+ {
- i.types[this_operand] |= BaseIndex;
- }
+ if (!i.base_reg)
+ i.base_reg = reg;
+ else if (!i.index_reg)
+ i.index_reg = reg;
+ else
+ {
+ as_bad (_("Too many register references in memory operand"));
+ return 0;
+ }
- /* Offset modifier. Add the register to the displacement string to be
- parsed as an immediate expression after we're done. */
- else if (intel_parser.op_modifier == OFFSET_FLAT)
- strcat (intel_parser.disp, reg->reg_name);
+ i.types[this_operand] |= BaseIndex;
+ }
- /* It's neither base nor index nor offset. */
- else
- {
- i.types[this_operand] |= reg->reg_type & ~BaseIndex;
- i.op[this_operand].regs = reg;
- i.reg_operands++;
- }
+ /* Offset modifier. Add the register to the displacement string to be
+ parsed as an immediate expression after we're done. */
+ else if (intel_parser.in_offset)
+ {
+ as_warn (_("Using register names in OFFSET expressions is deprecated"));
+ strcat (intel_parser.disp, reg->reg_name);
+ }
- /* Since registers are not part of the displacement string (except
- when we're parsing offset operands), we may need to remove any
- preceding '+' from the displacement string. */
- if (*intel_parser.disp != '\0'
- && intel_parser.op_modifier != OFFSET_FLAT)
- {
- char *s = intel_parser.disp;
- s += strlen (s) - 1;
- if (*s == '+')
- *s = '\0';
- }
+ /* It's neither base nor index nor offset. */
+ else if (!intel_parser.is_mem)
+ {
+ i.types[this_operand] |= reg->reg_type & ~BaseIndex;
+ i.op[this_operand].regs = reg;
+ i.reg_operands++;
+ }
+ else
+ {
+ as_bad (_("Invalid use of register"));
+ return 0;
+ }
- return 1;
- }
+ /* Since registers are not part of the displacement string (except
+ when we're parsing offset operands), we may need to remove any
+ preceding '+' from the displacement string. */
+ if (*intel_parser.disp != '\0'
+ && !intel_parser.in_offset)
+ {
+ char *s = intel_parser.disp;
+ s += strlen (s) - 1;
+ if (*s == '+')
+ *s = '\0';
+ }
- /* e11 id */
- else if (cur_token.code == T_ID)
- {
- /* Add the identifier to the displacement string. */
- strcat (intel_parser.disp, cur_token.str);
- intel_match_token (T_ID);
+ return 1;
+ }
- /* The identifier represents a memory reference only if it's not
- preceded by an offset modifier. */
- if (intel_parser.op_modifier != OFFSET_FLAT)
- intel_parser.is_mem = 1;
+ /* e11 BYTE
+ | WORD
+ | DWORD
+ | FWORD
+ | QWORD
+ | TBYTE
+ | OWORD
+ | XMMWORD */
+ case T_BYTE:
+ case T_WORD:
+ case T_DWORD:
+ case T_FWORD:
+ case T_QWORD:
+ case T_TBYTE:
+ case T_XMMWORD:
+ intel_match_token (cur_token.code);
- return 1;
- }
+ if (cur_token.code == T_PTR)
+ return 1;
- /* e11 constant */
- else if (cur_token.code == T_CONST
- || cur_token.code == '-'
- || cur_token.code == '+')
- {
- char *save_str;
+ /* It must have been an identifier. */
+ intel_putback_token ();
+ cur_token.code = T_ID;
+ /* FALLTHRU */
- /* Allow constants that start with `+' or `-'. */
- if (cur_token.code == '-' || cur_token.code == '+')
+ /* e11 id
+ | constant */
+ case T_ID:
+ if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
{
- strcat (intel_parser.disp, cur_token.str);
- intel_match_token (cur_token.code);
- if (cur_token.code != T_CONST)
- {
- as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
- cur_token.str);
- return 0;
- }
- }
+ symbolS *symbolP;
- save_str = (char *) malloc (strlen (cur_token.str) + 1);
- if (save_str == NULL)
- abort ();
- strcpy (save_str, cur_token.str);
+ /* The identifier represents a memory reference only if it's not
+ preceded by an offset modifier and if it's not an equate. */
+ symbolP = symbol_find(cur_token.str);
+ if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
+ intel_parser.is_mem = 1;
+ }
+ /* FALLTHRU */
- /* Get the next token to check for register scaling. */
- intel_match_token (cur_token.code);
+ case T_CONST:
+ case '-':
+ case '+':
+ {
+ char *save_str, sign = 0;
- /* Check if this constant is a scaling factor for an index register. */
- if (cur_token.code == '*')
- {
- if (intel_match_token ('*') && cur_token.code == T_REG)
- {
- if (!intel_parser.is_mem)
- {
- as_bad (_("Register scaling only allowed in memory operands."));
- return 0;
- }
+ /* Allow constants that start with `+' or `-'. */
+ if (cur_token.code == '-' || cur_token.code == '+')
+ {
+ sign = cur_token.code;
+ intel_match_token (cur_token.code);
+ if (cur_token.code != T_CONST)
+ {
+ as_bad (_("Syntax error: Expecting a constant, got `%s'"),
+ cur_token.str);
+ return 0;
+ }
+ }
- /* The constant is followed by `* reg', so it must be
- a valid scale. */
- if (strchr ("01248", *save_str))
- {
- i.index_reg = cur_token.reg;
- i.types[this_operand] |= BaseIndex;
-
- /* Set the scale after setting the register (otherwise,
- i386_scale will complain) */
- i386_scale (save_str);
- intel_match_token (T_REG);
-
- /* Since registers are not part of the displacement
- string, we may need to remove any preceding '+' from
- the displacement string. */
- if (*intel_parser.disp != '\0')
- {
- char *s = intel_parser.disp;
- s += strlen (s) - 1;
- if (*s == '+')
- *s = '\0';
- }
+ save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
+ strcpy (save_str + !!sign, cur_token.str);
+ if (sign)
+ *save_str = sign;
- free (save_str);
+ /* Get the next token to check for register scaling. */
+ intel_match_token (cur_token.code);
- return 1;
- }
- else
- return 0;
- }
+ /* Check if this constant is a scaling factor for an index register. */
+ if (cur_token.code == '*')
+ {
+ if (intel_match_token ('*') && cur_token.code == T_REG)
+ {
+ const reg_entry *reg = cur_token.reg;
+
+ if (!intel_parser.in_bracket)
+ {
+ as_bad (_("Register scaling only allowed in memory operands"));
+ return 0;
+ }
+
+ if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */
+ reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
+ else if (i.index_reg)
+ reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
+
+ /* The constant is followed by `* reg', so it must be
+ a valid scale. */
+ i.index_reg = reg;
+ i.types[this_operand] |= BaseIndex;
+
+ /* Set the scale after setting the register (otherwise,
+ i386_scale will complain) */
+ if (!i386_scale (save_str))
+ return 0;
+ intel_match_token (T_REG);
+
+ /* Since registers are not part of the displacement
+ string, we may need to remove any preceding '+' from
+ the displacement string. */
+ if (*intel_parser.disp != '\0')
+ {
+ char *s = intel_parser.disp;
+ s += strlen (s) - 1;
+ if (*s == '+')
+ *s = '\0';
+ }
+
+ free (save_str);
+
+ return 1;
+ }
- /* The constant was not used for register scaling. Since we have
- already consumed the token following `*' we now need to put it
- back in the stream. */
- else
+ /* The constant was not used for register scaling. Since we have
+ already consumed the token following `*' we now need to put it
+ back in the stream. */
intel_putback_token ();
- }
+ }
- /* Add the constant to the displacement string. */
- strcat (intel_parser.disp, save_str);
- free (save_str);
+ /* Add the constant to the displacement string. */
+ strcat (intel_parser.disp, save_str);
+ free (save_str);
- return 1;
+ return 1;
+ }
}
as_bad (_("Unrecognized token '%s'"), cur_token.str);
}
else
{
- as_bad (_("Unexpected token `%s'\n"), cur_token.str);
+ as_bad (_("Unexpected token `%s'"), cur_token.str);
return 0;
}
}
/* The new token cannot be larger than the remainder of the operand
string. */
- new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
- if (new_token.str == NULL)
- abort ();
+ new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
new_token.str[0] = '\0';
if (strchr ("0123456789", *intel_parser.op_string))
new_token.code = T_ID;
}
- else if (strchr ("+-/*:[]()", *intel_parser.op_string))
- {
- new_token.code = *intel_parser.op_string;
- new_token.str[0] = *intel_parser.op_string;
- new_token.str[1] = '\0';
- }
-
else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
&& ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
{
Otherwise, it's operator '.' followed by an expression. */
if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
{
- new_token.code = *q;
- new_token.str[0] = *q;
+ new_token.code = '.';
+ new_token.str[0] = '.';
new_token.str[1] = '\0';
}
else
*p++ = *q++;
*p = '\0';
- if (strcasecmp (new_token.str, "BYTE") == 0)
+ if (strcasecmp (new_token.str, "NOT") == 0)
+ new_token.code = '~';
+
+ else if (strcasecmp (new_token.str, "MOD") == 0)
+ new_token.code = '%';
+
+ else if (strcasecmp (new_token.str, "AND") == 0)
+ new_token.code = '&';
+
+ else if (strcasecmp (new_token.str, "OR") == 0)
+ new_token.code = '|';
+
+ else if (strcasecmp (new_token.str, "XOR") == 0)
+ new_token.code = '^';
+
+ else if (strcasecmp (new_token.str, "SHL") == 0)
+ new_token.code = T_SHL;
+
+ else if (strcasecmp (new_token.str, "SHR") == 0)
+ new_token.code = T_SHR;
+
+ else if (strcasecmp (new_token.str, "BYTE") == 0)
new_token.code = T_BYTE;
else if (strcasecmp (new_token.str, "WORD") == 0)
else if (strcasecmp (new_token.str, "DWORD") == 0)
new_token.code = T_DWORD;
+ else if (strcasecmp (new_token.str, "FWORD") == 0)
+ new_token.code = T_FWORD;
+
else if (strcasecmp (new_token.str, "QWORD") == 0)
new_token.code = T_QWORD;
- else if (strcasecmp (new_token.str, "XWORD") == 0)
- new_token.code = T_XWORD;
+ else if (strcasecmp (new_token.str, "TBYTE") == 0
+ /* XXX remove (gcc still uses it) */
+ || strcasecmp (new_token.str, "XWORD") == 0)
+ new_token.code = T_TBYTE;
+
+ else if (strcasecmp (new_token.str, "XMMWORD") == 0
+ || strcasecmp (new_token.str, "OWORD") == 0)
+ new_token.code = T_XMMWORD;
else if (strcasecmp (new_token.str, "PTR") == 0)
new_token.code = T_PTR;
/* ??? This is not mentioned in the MASM grammar. */
else if (strcasecmp (new_token.str, "FLAT") == 0)
- new_token.code = T_OFFSET;
+ {
+ new_token.code = T_OFFSET;
+ if (*q == ':')
+ strcat (new_token.str, ":");
+ else
+ as_bad (_("`:' expected"));
+ }
else
new_token.code = T_ID;
}
}
+ else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
+ {
+ new_token.code = *intel_parser.op_string;
+ new_token.str[0] = *intel_parser.op_string;
+ new_token.str[1] = '\0';
+ }
+
+ else if (strchr ("<>", *intel_parser.op_string)
+ && *intel_parser.op_string == *(intel_parser.op_string + 1))
+ {
+ new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
+ new_token.str[0] = *intel_parser.op_string;
+ new_token.str[1] = *intel_parser.op_string;
+ new_token.str[2] = '\0';
+ }
+
else
- as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
+ as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
intel_parser.op_string += strlen (new_token.str);
cur_token = new_token;
static void
intel_putback_token ()
{
- intel_parser.op_string -= strlen (cur_token.str);
- free (cur_token.str);
+ if (cur_token.code != T_NIL)
+ {
+ intel_parser.op_string -= strlen (cur_token.str);
+ free (cur_token.str);
+ }
cur_token = prev_token;
/* Forget prev_token. */
if (flag_code == CODE_64BIT)
{
regnames = regnames_64;
- regnames_count = sizeof (regnames_64);
+ regnames_count = ARRAY_SIZE (regnames_64);
}
else
{
regnames = regnames_32;
- regnames_count = sizeof (regnames_32);
+ regnames_count = ARRAY_SIZE (regnames_32);
}
for (regnum = 0; regnum < regnames_count; regnum++)
cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
}
+
+int
+i386_elf_section_type (const char *str, size_t len)
+{
+ if (flag_code == CODE_64BIT
+ && len == sizeof ("unwind") - 1
+ && strncmp (str, "unwind", 6) == 0)
+ return SHT_X86_64_UNWIND;
+
+ return -1;
+}
+
+#ifdef TE_PE
+void
+tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
+{
+ expressionS expr;
+
+ expr.X_op = O_secrel;
+ expr.X_add_symbol = symbol;
+ expr.X_add_number = 0;
+ emit_expr (&expr, size);
+}
+#endif